Streamed group offloading with `low_cpu_mem_usage=True` re-pins every group on every forward pass
Los mantenedores suelen responder en 1 día
Nadie ha tomado este issue todavía.
Evaluación
- Dificultad
- 4/5
- Tiempo estimado
- 3-5 días
- Aptitud para principiantes
- 42/100
- Tipo de issue
- Error
- Claridad
- Bastante claro
- Estado de actividad
- Activo
- Área
- machine-learning, performance
Línea de trabajo
Start with group_offloading.py, focusing on ModuleGroup._pinned_memory_tensors() and _apply_group_offloading_block_level; the issue also names tests/hooks/test_group_offloading.py. Trace how streamed groups pin CPU tensors and how unmatched_group handles offloading when low CPU memory usage is enabled. Done means reusable staging buffers avoid repeated pinning, original CPU tensors are restored on offload, and a test covers the behavior; the reported video difference still needs a same-process comparison.
Escrito por el modelo de indexación a partir del texto del issue.
Descripción
Problem
With streamed group offloading and low_cpu_mem_usage=True:
transformer.enable_group_offload(
onload_device=torch.device("cuda"),
offload_device=torch.device("cpu"),
offload_type="block_level",
num_blocks_per_group=1,
use_stream=True,
record_stream=True,
low_cpu_mem_usage=True,
)
ModuleGroup._pinned_memory_tensors() calls tensor.pin_memory() on every parameter of the group each time the group is onloaded, and drops the pinned copies right after the transfer. For a large transformer this means allocating and page-locking the full model size in host memory on every denoising step (LTX-2.5's transformer is ~38 GB in BF16, so 8 steps pin ~300 GB in total).
A second, smaller issue: the root unmatched_group built in _apply_group_offloading_block_level is created without low_cpu_mem_usage=config.low_cpu_mem_usage, and because it has stream=None its offload path does .to(offload_device), which allocates fresh CPU copies of those tensors instead of returning to the original (e.g. safetensors-mmapped) storage.
Proposed change
- Keep a small reusable pinned staging buffer per stream (two slots, each sized to the largest group, so the current transfer and the prefetched next group never share a slot). Contiguous unpinned CPU tensors are copied into a slot view and transferred from there; TorchAO and non-contiguous tensors keep the current path.
- When
low_cpu_mem_usage=True, remember the original CPU tensors and restore them on offload, and passlow_cpu_mem_usagethrough tounmatched_group.
Measurements
Lightricks/LTX-2.5-Diffusers, BF16, 512×320, 49 frames, 8 distilled steps, seed 42, prompt pre-encoded; connectors, transformer, VAE, audio VAE and vocoder offloaded with the settings above. AMD Radeon PRO W7800 (32 GB), torch 2.13.0+rocm10.0.0. Mean of two warm pipeline calls after one warmup call:
| Warm end-to-end | Process RssAnon |
|
|---|---|---|
main |
55.62 s (56.34, 54.89) | 10.56 GiB |
| proposed | 49.82 s (49.66, 49.97) | 9.18 GiB |
The audio output is bitwise identical between the two runs. The decoded video differs by up to 1.8e-2 (values in [0, 1]); the change only alters how identical bytes are staged on the host, so I suspect cross-process convolution algorithm selection in the VAE decode, but I have not isolated it yet and will confirm with a same-process comparison before opening a PR.
Coordination
This is separate from #12747 (configurable pinning of groups). If the approach looks reasonable, I can open a small PR (changes in group_offloading.py plus a test in tests/hooks/test_group_offloading.py).
- Lenguaje dominante
- Python
- Estrellas
- 34.9k
- Forks
- 7.5k
- Merge medio
- 3 d 16 h
- PR fusionados (30 d)
- 65
Preparar el entorno
- Sin Dockerfile ni archivo de Docker Compose
- Tiene una plantilla de pull request
- Leer la guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de huggingface/diffusers
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 84/100
huggingface/diffusers#15363 ·
Los mantenedores suelen responder en 1 día
-
`ZImageImg2ImgPipeline` / `ZImageInpaintPipeline`: explicit height/width different from the input image raises RuntimeErrorPosiblemente ocupada @chessing234 la tomó hace 1 día. Abiertobug
Dificultad 2/5 1-3 horas Aptitud para principiantes 70/100
huggingface/diffusers#15337 · 2 comentarios ·
Los mantenedores suelen responder en 1 día
-
Multi-vector textual inversion uses only the first vector when the token has uppercase letters (transformers v5)Posiblemente ocupada @sunhouy la tomó hace 2 días. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 75/100
huggingface/diffusers#15313 ·
Los mantenedores suelen responder en 1 día
-
TCDScheduler advances the generator on the final step of a strength-truncated custom schedulePosiblemente ocupada @SIDDARTHAREDDY8 la tomó hace 5 días. Abiertobug
Dificultad 2/5 1-3 horas Aptitud para principiantes 74/100
huggingface/diffusers#15288 · 2 comentarios ·
Los mantenedores suelen responder en 1 día
-
unload_textual_inversion(token) leaves the extra tokens of a multi-vector embedding behind, so it can't be loaded againPosiblemente ocupada @sohailshk la tomó hace 11 días. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 72/100
huggingface/diffusers#15211 · 3 comentarios ·
Los mantenedores suelen responder en 1 día
Todos los issues de huggingface/diffusers
Issues similares
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 73/100
Los mantenedores suelen responder en 1 día
-
Unused loot bag group names left in dreamcraft en_US.langPosiblemente ocupada Un pull request vinculado a esta issue está abierto o ya se fusionó. AbiertoStatus: Triage Suggestion
Dificultad 1/5 Menos de una hora Aptitud para principiantes 90/100
GTNewHorizons/GT-New-Horizons-Modpack#27789 ·
Los mantenedores suelen responder en 1 día
-
bug frontend sev-5
Dificultad 2/5 1-3 horas Aptitud para principiantes 83/100
-
core good first issue
Dificultad 2/5 1-3 horas Aptitud para principiantes 65/100
-
b4b code health next
Dificultad 2/5 1-3 horas Aptitud para principiantes 68/100
ESCOMP/CTSM#4330 · 1 comentario ·
Los mantenedores suelen responder en 1 día